Titre : Privacy-preserving Machine Learning Techniques
Date : 10 décembre 2021
Résumé : In this thesis, we aim to design such protocols for Machine Learning as a Service to compute neural network classification, TRAjectory CLUStering, and data aggregation under privacy protection, namely: PAC, SwaNN, ProteiNN, pp-TRACLUS, and PRIDA. In these solutions, our goal is to guarantee data privacy as well as to provide accurate/qualified clustering and efficient evaluations when executing such protocols. In order to preserve data privacy, we employ several cryptographic techniques: Secure two-party computation, homomorphic encryption, and homomorphic proxy re-encryption. We utilise several optimisation techniques such as packing to lessen the computational cost and bandwidth overhead to obtain efficient protocols. Our solutions show promising performance results and call for future work.
Lieu : à EURECOM et par Zoom